feat: add gemini provider

[fabienpiette]

Summary

• Add Google Gemini as an alternative AI provider for both transcription and post-processing, alongside the existing OpenAI support
• Implement automatic provider fallback — when both API keys are configured, goscribe switches to the alternate provider on failure (rate limits, API errors)
• Add rate limit handling with exponential backoff and inter-chunk delays

Changes

New Gemini provider (main.go):

• Native Gemini audio transcription via inline base64-encoded audio data
• Gemini text generation for post-processing actions
• MIME type detection for audio files (mp3, wav, ogg, flac, aac, etc.)
• Gemini-specific context limits (up to 900K tokens for Gemini 1.5/2.0)
• 20MB file size limit for Gemini (vs 25MB for OpenAI)

Provider management:

• -provider flag to select openai or gemini per invocation
• -set-provider to persist the default in config
• -gemini-key / -set-gemini-key for Gemini API key management
• -no-fallback flag to disable automatic provider switching

Resilience:

• Exponential backoff with retry on rate-limited requests
• Parses retry-after from API error responses
• Inter-chunk delays to prevent hitting rate limits during large file processing

Tests (main_test.go):

• 475 new lines of tests covering provider selection, Gemini key storage, MIME type detection, config validation, and model context limits

Docs (README.md):

• Updated with dual-provider usage, new CLI options table, provider comparison, fallback behavior, and rate limit documentation

Test plan

• Run make test to verify all unit tests pass
• Test transcription with OpenAI provider (goscribe meeting.mp3)
• Test transcription with Gemini provider (goscribe -provider gemini meeting.mp3)
• Test fallback behavior with an invalid OpenAI key but valid Gemini key
• Test -no-fallback correctly disables provider switching
• Test -set-gemini-key and -set-provider persist to config